Commits on Apr 24, 2018

1. Rewrite the mpsc channels …

   Previously, the bounded mpsc channels wouldn't provide backpressure
   in many cases because they provided a guaranteed slot for every
   sender. This change refactors mpsc channels to never allow more than
   buffer + 1 messages in-flight at a time.
   
   Senders will attempt to increment the number of in-flight messages
   in poll-ready. If successful, the count is incremented and a message
   is sent. If the buffer was full, the sender adds itself to a queue
   to be awoken. When a message is received, the receiver attempts to
   awaken a Sender and grant it "sending rights". If no such channel
   was found, it decreases the number of in-flight messages currently
   present in the buffer.
   
   There's one odd choice in this change, which was to pick buffer + 1
   rather than buffer as the max number of elements. In practice, many
   of the tests and a handful of real-world code relied on the ability
   to have an initial element available for the first sender, and were
   attempting to send into channels with a buffer size of zero. These
   use-cases were all broken by this change without the change to use
   buffer + 1.
   
   Despite the buffer + 1 mitigation, this is still a breaking change
   since it modifies the maximum number of elements allowed into a
   channel. 0.2 has been out for a short enough time that it's probably
   possible to transition without breaking existing code, but this
   approach should be taken with caution. Luckily, 0.2 organized
   futures-channel into a separate crate, so it could be released as
   an 0.3 version of futures-channel, futures-sink, futures-util, and
   the futures facade while still maintaining futures-io and futures-core
   compatibility.

   

   cramertj committed Apr 24, 2018
   Configuration menu

   • View commit details
   • Copy full SHA for d732372
   • Browse repository at this point

   Copy the full SHA

   d732372 View commit details

   Browse the repository at this point in the history